Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.37530

Back-End-Of-Line Integration of 2-Terminal Ferroelectric Hafnia Synaptic Weights on Commercial CMOS Process

Alexander Flasby·Anwesha Panda·Nathan Savoia·Laura Bégon-Lours
View original source ↗Read source document ↗Export metadata ↓

Abstract

Functionalising the back-end-of-line (BEOL) of CMOS with emerging materials is a key route to neural network accelerators and neuromorphic systems. In this work, we demonstrate for the first time the integration of 2-terminal ferroelectric synaptic weights in the BEOL of as-delivered commercial 180 nm CMOS dies. A novel material stack is proposed, in which a WOx/[HfO2-ZrO2] bilayer is sandwiched between TiN and Mo electrodes. For devices annealed at 450 °C, programming below 1.8 V is achieved with a cycle-to-cycle variation below 3 %. Synaptic weights annealed at 500 °C exhibit more gradual weight updates, while the tested underlying NMOS transistors remain functional after the complete integration process. The trade-off between thermal budget and on-device training performance is illustrated using hardware-aware neural network simulations.

Explore related subjects

cond-mat.mtrl-sciphysics.app-ph

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Alexander Flasby, Anwesha Panda, Nathan Savoia, Laura Bégon-Lours. 2026-10-27. Back-End-Of-Line Integration of 2-Terminal Ferroelectric Hafnia Synaptic Weights on Commercial CMOS Process. https://arxiv.org/abs/2610.37530

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Dissolution-induced mode transition of crack propagation in carbonate rocks: From brittle fracture to diffuse damage

In chemically reactive environments, the mechanical integrity of geomaterials is fundamentally compromised by solid matrix dissolution. In this study, we propose a fully coupled chemo-mechanical phase-field framework to capture the dynamic interplay between mineral dissolution and fracture propagation. A key feature of the proposed model is the dynamic coupling of local mass removal to the fracture length scale, while also incorporating the damage-accelerated reaction-diffusion processes. Our results capture the development of an enlarged fracture process zone driven by chemical mass removal. This chemically induced widening blunts the sharp crack tip, alleviating the near-tip stress concentrations and causing a pronounced degradation in material stiffness before failure. Furthermore, we reveal a distinct ductilization effect, characterized by a more gradual accumulation of damage and a delayed onset of macroscopic failure. We show that the transition between brittle and ductile failure modes is dictated by the competing timescales of chemical degradation and mechanical deformation. Highly acidic environments enhance matrix dissolution and promote ductile fracture, whereas rapid mechanical loading limits chemical interaction and preserves brittle failure mode.

cond-mat.mtrl-sci↗
arXiv2026

Sustainability-Aware Discovery of Dual-Atom Hydrogen Evolution Catalysts via a Unified Machine Learning Framework for Activity and Carbon Cost

Catalyst discovery for green hydrogen has optimized activity alone, deferring economic and environmental costs to stages where they disqualify otherwise promising candidates. Here we introduce a unified machine learning framework that screens catalytic activity and monetized carbon cost in a single pipeline. A machine learning potential predicts hydrogen evolution activity and stability at density-functional-theory accuracy. A second model, trained on published estimates of the social cost of carbon, converts each candidate's embodied CO2 into a monetary burden. Screening 460 graphene-supported dual-atom catalysts across 23 transition metals and 20 nitrogen-coordination motifs yields 26 active, synthesizable candidates. The two axes then conflict. The most active candidates span Mn-, Fe-, Zr-, Cr-, Cu-, Ni-, Ti-, Nb-, V-, Pd-, Au-, and Rh-based catalysts, indistinguishable in adsorption energy yet spanning a 58,000-fold range in carbon cost. Joint optimization selects Mn2@2Nf and Zr2@2Nc, matching the best precious-metal activity at tens-of-thousands-fold lower carbon cost, independent of catalyst loading.

cond-mat.mtrl-sci↗
arXiv2026

Crystal forming ability of amorphous refractory metals under nanoindentation: a molecular dynamics study

Amorphous refractory metal coatings combine high hardness with chemical inertness, yet their metastability makes them prone to mechanically induced crystallisation (devitrification) under contact loading, and how readily such a glass re-orders to its parent body centred cubic (bcc) crystal is unknown across the refractory series. We prepared amorphous V, Nb, Mo, Ta and W by melt quenching to 300 K, benchmarked each interatomic potential against ab initio liquid radial distribution functions and self-diffusion coefficients, and probed them by large scale molecular dynamics nanoindentation. Indentation drives a localised amorphous to bcc transformation by bulk nucleation, growth and coalescence. A crystal forming ability (CFA), the maximum slope of the sigmoidal bcc fraction versus depth curve, spans about a factor of four and decreases as V > Mo > Nb > Ta > W; it falls with indenter velocity as CFA scales as v^(-m), with a mean exponent of 1.08, so CFA.v is nearly constant and the transformation rate is almost velocity independent. Transforming atoms carry excess non affine displacement and local shear strain, not hydrostatic pressure, marking a shear associated displacive pathway; the bulk driving force is largest for the most resistant element, W, so resistance tracks cohesive bond strength rather than the thermodynamic driving force. Early bcc like order correlates with the sharpness and depth of the transition, whereas the mechanical work to half transformation correlates with the peak density of persistent bcc clusters. A cluster population balance links nucleation, growth and coalescence to a terminal microstructure whose completeness does not follow the CFA order.

cond-mat.mtrl-sci↗